Graph-Based Community Detection for Identifying Misinformation Propagation Patterns on Social Media

Meilani, Budanis Dwi and Noche, Elmar B. and Cruz, Engelbert Q. (2026) Graph-Based Community Detection for Identifying Misinformation Propagation Patterns on Social Media. Journal of Data Science, 2026 (19). pp. 323-343. ISSN 2805-5160

[img] Text
jods2026_19.pdf - Published Version
Available under License Creative Commons Attribution.

Download (221kB)
[img] Text
966 - Published Version
Available under License Creative Commons Attribution.

Download (34kB)
Official URL: http://ipublishing.intimal.edu.my/jods.html

Abstract

The rapid spread of misinformation through social media networks poses significant challenges to information quality, public trust, and digital security operations. Traditional content-based misinformation detection approaches are useful for analysing textual signals, but they often underrepresent the relational mechanisms through which misleading narratives are amplified, clustered, and transferred between users. This study develops a graph-based community detection framework for identifying misinformation propagation patterns in social media interaction networks. User interactions are transformed into weighted graph structures in which nodes represent accounts and edges represent reshare, reply, mention, or co-engagement relationships. The Louvain algorithm is applied for modularity-optimised community discovery, and the resulting communities are evaluated against baseline graph clustering methods using modularity, coverage, conductance, normalized mutual information, partition stability, and propagation-risk indicators. Centrality analysis using degree, betweenness, eigenvector centrality, and PageRank is further used to identify influential bridge accounts and potential propagation bottlenecks. The experimental graph contains 25,104 user nodes, 169,312 interaction edges, and 2,418 information cascades across a 30-day observation window. Results show that the Louvain model achieves the highest modularity score of 0.501 and the best balance between coverage and conductance. Misinformation-dominant communities display higher internal density, stronger centralisation around a smaller group of bridge accounts, and deeper propagation pathways than verified-information communities. The top 10% of bridge accounts explain 38.7% of cross-community misinformation diffusion, indicating that structural monitoring can support early-risk prioritisation. The study contributes a reproducible graph mining framework for information-quality monitoring, misinformation intelligence, and digital trust enhancement in social media environments.

Item Type: Article
Uncontrolled Keywords: Graph mining; Social network analysis; Community detection; Misinformation; Data analytics; Freedom of information
Subjects: H Social Sciences > H Social Sciences (General)
Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76 Computer software
Depositing User: Unnamed user with email masilah.mansor@newinti.edu.my
Date Deposited: 18 Nov 2026 08:27
Last Modified: 18 Nov 2026 08:27
URI: http://eprints.intimal.edu.my/id/eprint/2390

Actions (login required)

View Item View Item